Magnitude and Performance of Percentage Body Fat and Fat Mass Index in Determining Overweight/Obesity among University Undergraduate Students in the Kumasi Metropolis, Ghana.

Background: Anthropometric indices such as body mass index (BMI) is a widely used index of adiposity across clinical settings, yet there are limitations regarding their estimates of body fat. We determined the magnitude and the performance of fat mass index (FMI), and percentage body fat (PBF) for determining obesity among undergraduate students in the Kumasi Metropolis, Ghana. Methods: This University community-based cross-sectional study included 1,552 undergraduate students (700 males and 852 females) aged 15–29 years from the Kwame Nkrumah University of Science and Technology, Kumasi, Ghana. Anthropometric measurements and bioelectrical impedance analyses were performed. Data on sociodemographic and lifestyle-related characteristics were obtained using a structured questionnaire. Results: Comparing male to female students, the age-standardised prevalence of overweight/obesity was 10.8%/3.5% vs. 19.3%/7.6% (p<0.0001) for BMI criterion, 16.4%/8.5% vs. 21.9%/19.9% (p<0.0001) for PBF criterion and 16.0%/9.2 vs. 21.2%/22.0% for FMI criterion (p<0.0001), respectively. There was a stronger agreement between PBF and FMI in the classication of overweight and obesity in terms of males (kappa=0.855) and females (kappa=0.865) but a weak agreement between BMI and PBF and between BMI and FMI. About 29.3% of males and 61.4% of the females categorized as normal weight by BMI were identied as overweight by PBF. Similarly, 3.0% of male and 1.3% of females dened as normal weight by BMI were obese when examined by PBF. FMI can be predicted from BMI and BMI 2 with much accuracy (mean bias = -0.1). Conclusion: The inconsistency observed between BMI and PBF reects a limitation of BMI. FMI accurately assessed obesity with many conveniences as BMI to the same extent.


Introduction
Over the past three decades, chronic conditions such as diabetes, cardiovascular conditions, and obesity accounted for <45% of the global disease burden [1]. Today, obesity, which represents an abnormal or excessive fat accumulation, is one of the major public health problems leading the toll on death and disability worldwide [1]. Prevalence of obesity across the world has tripled (>100% increase) since 1975 with nearly 1.9 billion adults estimated to be overweight in 2016, including 650 million adults with obesity [2]. Especially in developing countries, the possible implications of obesity on current and future population health and health care spending is likely to be blatantly enormous [3]. The implication of this is that health bene ts encouraged by modern medicine are being eroded by the current obesity epidemic. In Ghana, overweight and obesity prevalence is estimated to be 25.4% and 17.1%, respectively which are signi cantly higher in women than men and the patterns generally mimic the levels of urbanization [4].
Although there are simple and easily estimated measures of adiposity, BMI has been widely used as a marker of adiposity in both epidemiological and clinical practices [5,6]. The usefulness of BMI, however, is limited in identifying differences in body composition and body fat distribution. The categorization of overweight/obesity is broad and imprecise, especially when using BMI classi cation [5,7]. Regarding the effects of visceral fat accumulation on an increased risk of metabolic diseases, waist circumference (WC) [8], waist-to-height ratio (WHtR) [9], and body adiposity index [10] has been promising alternatives for estimating obesity. Also, bioelectrical impedance analysis (BIA) has been reliably established as a measure of body adiposity with equivalent sensitivity as underwater weighing (UWW) and dual-energy Xray absorptiometry (DEXA) [11,12]. Although BIA is widely used in clinical and epidemiologic eld [11,12], it requires predictive equations for the determination of fat-free mass that are speci cally developed for different populations, ethnicities, age groups, and sex [13]. Thus, there is still a need for an even more simpli ed approach to assessing body fat content in large epidemiological studies and the clinical setting.
A potential indicator of body adiposity is fat mass index (FMI). This index since its introduction in 1990 [14], has been reliably applied in several populations [15,16]. It is de ned by taking the body fat mass component from BIA and dividing by height squared. Although FMI, BMI and PBFmaybe composed of the same variables and similar, each one of them, has been shown to categorize overweight and obesity differently [12,16]. BMI takes into account body weight and body height while FMI requires information on body weight, body height, and fat mass content. Comparing FMI to other indices, this index is highly in uenced by fat mass content, which is the desired variable to be measured in obesity studies [17]. Also, the superiority of FMI over PBF has been shown in some studies [14,15].
This study focused on describing the prevalence of overweight/obesity among tertiary education students in the Kumasi metropolis and proposed FMI cut-off points for this population. Additionally, we examined the performance of BMI, WC, WHtR, AVI, and BAI, and developed a predictive model of FMI that can be used in eld studies requiring body composition assessment in Ghanaian young adults without the need for sophisticated equipment or techniques subject to operator error. This approach is useful for weight management programs especially among the youths that are actively involved in bodybuilding activities to identify appropriate indices de ning weight-related behaviours.

Study design/setting
This cross-sectional study was undertaken at the Kwame Nkrumah University of Science and Technology (KNUST), Kumasi, from August 2018 to July 2019.

Subject selection
A multi-stage random sampling strati ed was used to select 1552 rst to fourth-year undergraduate students aged 15-29 years. These students were selected to cover the six Faculties in KNUST including the College of Art and Built, Social Sciences, Health Sciences, Engineering, Applied Sciences, Architecture, and Agricultural sciences. Students with any underlying conditions which are associated with abnormal weight were excluded. A structured questionnaire was used to collect data on socio-demographic characteristics, lifestyle risk factors (drinking of alcohol and smoking), and family history of obesity.

Sample size determination
To obtain a representative sample size for the entire student's population at all college level, at a con dence level of 97% and a margin of error of 0.05, the minimum sample size required for the study was 1090 using the formula below: Where N = total undergraduate student population (56,000), n = estimated sample size, e = margin of error. However, to ensure a fair distribution of the samples at the college level and stronger statistical power and effect size, the samples were projected to 1552 students.

Anthropometric measurement
Each participant stood up-straight on the main unit, looking straight, barefooted and with arms horizontally raised holding the display unit, extended at a 90° angle, and weight in kilograms (kg) was estimated using the Omron B511. All students stood straight, with feet placed together and at on the ground and a portable height rod Stadiometer was used to measure the body height. Waist circumference (WC) and Hip circumference (HC) were measured using a tape measure at the point of the umbilicus and maximum gluteal protrusion, respectively. PBF was estimated using the Omron B511. Omron B511 is a clinically validated full body composition monitor with 8 high-precision sensors for hand-to-foot measurement (OMRON HEALTHCARE Co., Ltd.). The machine conforms to EN60601-1-2:2015 Electro Magnetic Compatibility (EMC) standard and uses the bioelectrical impedance, along with height, weight, age and gender information to generate results based on OMRON's data of body composition [18]. PBF was expressed as the ratio of body fat mass (kg) to body weight (kg) expressed as a percentage. FMI was also estimated as the ratio of body fat mass (kg) to height (m) squared.

Calculated anthropometric indices
BMI (kg/m 2 ) was estimated from the ratio of weight to height (in meters) squared. WHtR and WHR were estimated as a ratio of WC to height and hip, respectively. BAI and AVI were estimated using the formulae below: De nitions of overweight/obesity Subjects were categorised as overweight if BMI was in the range 25.0-29.9 kg/m 2 , and obese if BMI ≥30.0 kg/m 2 [2]. Classi cation for overweight and obesity according to PBF criteria is shown in Table S1. Values for WC for overweight and Obesity were 94.0-101.9 and ≥102.0 cm for males; 80.0-87.9 and ≥88.0 cm for females, respectively. ROC analyses [ Figure S1] were performed to generate FMI cut-offs for male and female students; overweight (male = 4.5-6.5 Kg/m 2 , female = 7.4-9.8 Kg/m 2 ); Obesity (male >6.5 Kg/m 2 , female >9.8 Kg/m 2 ), respectively.

Statistical analyses
Prevalence of underweight, overweight, and obesity was standardised by age and sex. The correlation of WC, BMI and BAI with PBF and body fat mass (BFM) were investigated using linear regression models with Pearson's correlation coe cient (r), and the agreement was assessed by intraclass correlation coe cient (ICC). Multiple linear regression analysis was performed with FMI as the dependent variable to create predictive models of FMI for university undergraduate students. The following independent variables were assessed: age, WC, WHtR, BMI, and BMI 2 . Analyses were performed separately for male and female students because of the higher BFM at any given height seen in female compared to male students. The coe cients of determination (R 2 ) were reported as a measure of the proportion of the variability of FMI explained by the independent variables. To test whether the equations adequately predicted FMI in our entire study sample, we randomly split the dataset into two to (a) derive predictive equations in one dataset (N = 985) and (b) predict FMI in the other dataset (N = 567). In general, the following regression coe cients were obtained: 0.7-1.0, for strong correlation; 0.4-0.7, for a slight correlation. In the case of ICC evaluation score, 0.20 was considered slight, 0.21-0.40, fairer; 0.41-0.60, moderate; 0.61-0.80, substantial; and 0.81, almost perfect. A two-sided p-value of 0.05 was considered statistically signi cant. Statistical analyses were performed using MedCalc Software for Windows, version 18.91 (https://www.medcalc.org/).

Page 7/18
The descriptive summary of demographic and anthropometric data among male and female students is shown in Table 1. Male students were older than female students (mean age = 20.6 vs 19.8 years, p-value <0.0001). More males than females (p-value <0.0001) occasionally exercises (48.1% vs 37.8%) or exercises almost every day (16.3% vs 11.5%). The percentage of alcohol drinkers was higher in males (11.3%) compared with females (5.6%, p-value <0.0001). Also, a positive family history of diabetes was seen more in females than males (22.4% vs 17.3%, p-value = 0.012). The mean values of BAI, BMI, PBF, and BFM were higher among females (p-value <0.0001) whereas WC was higher in males than females (p-value <0.0001).    There was almost perfect agreement between FMI criteria and PBF criteria for de ning overweight and obesity among males (Kappa = 0.855) and female (kappa = 0.865). The percentage of men classi ed by BMI as being normal body weight and as overweight but classi ed as obese by PBF was 3.0% and 29.3%, respectively. In women, 1.3% and 61.4% of those classi ed by BMI as being normal body weight and as overweight, respectively, were found to be obese by PBF. When comparing BMI and FMI categories, 1.8% and 42.7% of the men in the normal and overweight BMI categories, respectively, were obese by FMI, whereas 61.4% of females in the overweight BMI category were obese by FMI. Poor agreement in overweight and obesity classi cation was observed between WC and PBF (kappa = 0.09) and FMI (kappa = 0.082) in males. Sex was identi ed as an effect modi er in the relationship of BMI with PBF and FMI [ Figure 1A]. BMI explained 91% variability in FMI among female participants and 77% in male participants. The proportional variability in PBF by BMI was lesser in females compared males (61% vs. 74%). BMI shows a slightly curvilinear relationship with PBF [ Figure 1D], and a more linear relationship with FMI in a secondorder polynomial regression analysis [ Figure 1C]. The curvilinear relationship between BMI and PBF is more accentuated in females than in males (R 2 = 0.78, p < 0.0001).
From for male students and FMI = -11.229 + 0.877 (BMI) -0.002 (BMI 2 ) for female student. Adding age, WC, and WHtR to the model marginally improved the model's predictability. A Bland-Altman plot comparing measured and predicted FMI for both male and female students shows good agreement between the methods with a mean difference of -0.10 [ Figure 2].

Discussion
Using the internationally accepted threshold for BMI, we observed an overweight prevalence of 11.0% among male students and 22.1% among female students. Further, 3.9% of male students and 7.6% of female students were obese. These gures were doubled when overweight/obesity prevalence was de ned by PBF. The prevalence estimates of overweight and obesity observed in this study are similar to those observed among University students from Botswana [20] and Ghana [21] who used the internationally accepted threshold for BMI. Also, Peltzer, et al. [22] in a cross-sectional study among university students from 22 countries reported that the prevalence rates of overweight and obesity were 18.9% and 5.8% among male students; 14.1% and 5.2% among female students, respectively which is similar our present ndings. Currently tertiary enrolment rate in Ghana stands at 16.2% [23], and overweight/obesity rates exceeding 10% re ect a signi cant national obesity problem. This rate may be biased towards high-income groups. This present study did not record data on the socio-economic status, detailed physical activity, and calorie intake by the students, which limits the study's ability to make conclusions regarding the lifestyle and dietary habits of students. However, we observed that femaleparticipants between the ages of 20-29 years, those with family history of obesity, and being on diet were signi cant factors for overweight/obesity prevalence among undergraduate university students in the Kumasi Metropolis.The nding is consistent with reports from other related studies [20,22].
The ability to estimate or quantify fat stores accurately is central to the prevention and treatment of obesity-related conditions. Out of the commonest methods used, we selected BMI, WC, WHtR, BAI, AVI, WHR, and BIA for comparison in a sample of university undergraduate students. Results from the present demonstrated a strong correlation between BMI and PBF as estimated by BIA. However, 29.3% of males and 61.4% of the females categorized as as normal weight by BMI were identi ed as overweight by PBF.
Similarly, 3.0% of male and 1.3% of females de ned as normal weight by BMI were obese when examined by BIA. This indicate that BMI has the limitation to differentiate between free fat mass and fat mass. Similar ndings had been reported by Pasco, et al. [5] and Pasco, et al. [7] among the general population. This nding re ects the phenomenon that increasing BMI only accurately mirrors increased weight-forheight but mask changes in body composition [5]. In a study by Peltz, et al. [15] 46.2%% of the men and 50.7% of the women with a BMI below 30 were found to be obese by PBF criteria when analysed by BIA.
Our results are in concordant with reports from other related studies that have provided evidence of the burden of obesity among the young adult population [4,20,22] and the limitation of BMI in de ning obesity despite its wider applicability in clinical and epidemiological studies. To overcome these limitations, some studies [5,7] have suggested lower cut-offs for BMI by sex and speci c ethnic groups.
Our results were not different as we observed a strong correlation but weak inter-rater agreement between BMI and PBF in both male and female students. Similar ndings have also been reported among different population groups [5,15,24,25].  [16,26,27].
The inconsistency observed between BMI and PBF re ects the major limitation of BMI in assessing adiposity. That is, BMI among the young adult population cannot be completely relied upon as a measure of adiposity. Therefore, its limitations must be taken into account when interpreting body weight classi cations based on BMI. FMI appears to provide not only accuracy but an economical advantage together with convenience for assessing obesity.
The study had some limitations. First, the BAI principle and device used in the study was limited by overestimating body fat in lean individuals and underestimating in very obese individuals. Second, we did not do a comparison to reference methods for the assessment of body adiposity. Nevertheless, the strength of the study lies in the large sample size, and the sampling technique employed, reducing bias in our ndings. Notably,the usefulness of FMI has not been explored in the Ghanaian population and not fully explored in literature, possibly due to the lack of speci c cut-offs [15]. So far, BMI is the simplest and most convenient method of assessing body adiposity. However, since BMI is mathematically equal to the sum of FMI and fat-free mass index according to studies [16,17], the use of FMI as a measure of adiposity is more appropriate and relates to the adiposity content of BMI.

Conclusion
Our study highlighted that the prevalence of overweight/obesity is high among university undergraduate students in the Kumasi Metropolis. Secondary, we highlighted that BMI is not reliable or su cient for identifying overweight/obesity. This suggest that de ning the incidence or prevalence of obesity based on BMI is likely to be substantively biased. PBF is also superior to BMI in correctly classifying obesity based on accurate estimates of body fatness. FMI is as convenient to use as BMI and PBF for de ning overweight/obesity. Moreover, FMI can be predicted from BMI and BMI 2 with much accuracy.

Declarations
Ethical Approval and Consent to Participate Permission to carry out the study on the university campus was obtained from the university authorities.
Written informed consent were obtained from all participants who opted to participate.

Competing interests
The authors declare that they have no competing interests.

Figure 1
Scatter plot of BMI against PBF and FMI for male and female Ghanaian University undergraduate students. Predicted values are represented by lines.